1. What Part II Was Looking For

An investigation does not always begin with a clear account of everything it will need.

Sometimes a problem can be stated before all of the distinctions on which its examination will depend have themselves become objects of attention. As inquiry proceeds, something that had remained in the background may become visible through the work it is already doing.

At that point we may know more clearly that something needs to be distinguished than we know exactly how the distinction should be understood.

Much of Conviction Formation Theory developed in this way.

The claim that becoming convinced is not under direct voluntary control opened the investigation, but following it repeatedly brought further distinctions and dependencies into view: orientation, the role of questions and possibilities, criteria, externalization, and participation. Recursive examination arose when the criteria and methods used in inquiry themselves became part of the problem.

Making something explicit does not settle its standing. What appeared important at first may later need qualification, division, or replacement. A representation can preserve the wrong thing, or omit something that later turns out to matter.

Still, inquiry need not wait until all of this has been settled. Sometimes the next step is to make clearer what is already carrying part of the inquiry, so that more can be asked of it.

Part II began from one such problem left by the first cycle.

We can examine convictions. We can ask what supports them, consider alternatives, compare evidence, or introduce criteria and procedures. Part I had already found that criteria can become sufficiently settled to organize inquiry, sometimes even between people whose larger orientations differ.

But those criteria remain inside conviction formation.

Logical validity has to acquire force for us, just as evidence or predictive success does. A criterion may organize an inquiry for years and later become questionable. Different criteria may also continue to matter while pulling an inquiry in different directions.

Recursive examination made this difficulty explicit.

A conviction may depend on an interpretation, an assumption, or a criterion that had remained in the background. Making such dependencies more explicit can expose further questions. And whatever we use to examine them can itself become an object of inquiry.

But a further question may be posed without becoming genuinely puzzling. We can examine a criterion while it continues to organize the inquiry. Usually only part of an orientation becomes unsettled while much of the rest continues to carry what happens next.

That gives inquiry enough local stability to continue, even while further examination remains possible.

What it does not yet give us is a criterion for assessing what happens through that examination.

Suppose an inquiry has exposed more of what a conviction depends on. Suppose a criterion has been applied carefully and an objection reconstructed. On what basis could we say that the conviction or the inquiry was now better, without simply introducing another criterion whose standing raised the same question?

The earlier notion of firmer ground gave only a partial answer. A conviction may survive examination that it had not undergone before. That can matter. It does not put different convictions on one general scale.

Part II turned to cases in which inquiry appeared to be strongly constrained while the conditions carrying that constraint could still be examined. Maybe something could be learned from them for our problem.

Logic, probability theory and Bayesian updating, and Pearl's theory of causality offered examples of such cases.

Logical conclusions can appear necessary. Probability constrains relations among possibilities even when several remain open. Causal models distinguish observed association from intervention.

These seemed like good places to look for constraint whose source could be reconstructed rather than merely asserted.

The strongest result would have been an independent criterion, something capable of governing inquiry without depending in turn on commitments that remained inside conviction formation.

Part II did not find such an independent criterion.

Logic required premises, interpretations, mappings, and a representation of the problem. Probability depended on a represented possibility space and on how weight was assigned within it. Causal reasoning depended on assumptions about causal structure and on how the situation was represented.

Any of these conditions could themselves become questionable.

But while enough of them remained convincing for the inquiry at hand, they could still constrain thinking.